Terahertz optical pattern recognition with rotation and scaling enhanced by a 3D-printed diffractive deep neural...

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Terahertz Technology

Background:

• Optical pattern recognition (OPR) offers cost-effective image recognition for terahertz (THz) imaging via single-point detection.
• Traditional OPR systems struggle with recognition spot offsets due to input image rotation and scaling.

Purpose of the Study:

• To develop a robust OPR method for THz imaging that overcomes limitations of rotation and scaling.
• To demonstrate a stable recognition spot irrespective of input image transformations.

Main Methods:

• Implementation of a full-diffractive method using an all-optical diffractive deep neural network.
• Construction of the network with two layers of diffractive optical elements (DOEs), integrated into a compact, 3D-printed all-in-one device without a 4f-system.

Main Results:

• Experimental validation of stable image recognition performance across 0° to 360° rotation.
• Demonstration of consistent recognition with image scaling ratios from 1 to 1/1.9.
• The diffractive deep neural network successfully maintained a fixed recognition spot under varying input conditions.

Conclusions:

• The developed all-optical diffractive deep neural network provides a robust solution for OPR in THz imaging.
• This technology enhances functionality for compact THz systems, particularly in imaging and security applications.
• The method's stability against rotation and scaling opens new possibilities for advanced THz sensing.